Run the LiveView test files in parallel closes #611 #616

Merged

simon merged 5 commits from issue/mitgliederverwaltung-611 into main

2026-10-20 23:37:17 +02:00

simon commented

2026-10-20 23:23:17 +02:00

Owner

closes #611

Description of the implemented changes

The changes were:

Bugfixing
New Feature
Breaking Change
Refactoring

Run the LiveView test files in parallel: most of the remaining serial block of the fast suite were LiveView test modules whose only serial reason was "flip deferred". They are now hardened against other tests' rows and host load and run async; every module that stays serial names its actual reason.

What has been changed?

State that the async log-capture guard covers the capture_log tag forms
Move the wall-clock and log-capture tests into their own serial modules
Keep LiveView tests independent of other tests' rows and of host load
Run the LiveView test modules async and name why the rest stay serial
Record the suite times after the LiveView flips and the async pitfalls they surfaced

Definition of Done

Code Quality

No new technical depths
Linting passed
Documentation is added were needed

Accessibility

New elements are properly defined with html-tags
Colour contrast follows WCAG criteria
Aria labels are added when needed
Everything is accessible by keyboard
Tab-Order is comprehensible
All interactive elements have a visible focus

Testing

Tests for new code are written
All tests pass
axe-core dev tools show no critical or major issues

Additional Notes

Fast suite (just test-fast) on the same 4 vCPU host, seeds 101/202/303, no other test run on the host, medians:

	Before	After	Δ
Total	227.4 s	162.8 s	−64.6 s (−28 %)
Async	55.9 s	74.6 s	+18.7 s
Sync	169.6 s	88.2 s	−81.4 s

30 test modules moved from serial to async (74.8 s of serial time). Every flipped module passed --repeat-until-failure 5 on its own and five seeded full-suite runs per batch, with no Postgres deadlock and no lock wait of 1 s or more. Under parallel load from another worktree on the same host the suite takes 180–215 s.

Modules that stay serial after the async proof: index_sort_atom_property_test (asserts the VM-wide atom count), index_value_control_test, index_bulk_scope_test, index_filter_chips_test, index_filter_input_hardening_test (assert_patch timed out under heavy host load; the hardening is kept), and member_field_live/index_component_test (a lock wait on the settings row; settings isolation is #540's scope). Re-flipping the four assert_patch modules with a longer patch timeout is a follow-up.

Deliberate deviation: six overview test modules of the same assert_patch class stay async with a 1 s patch timeout instead of going back to serial; they never failed themselves and held five full runs under heavy load.

No product code changes; the diff is limited to test/ and docs/.

Larger commits: "Keep LiveView tests independent of other tests' rows and of host load" and "Run the LiveView test modules async and name why the rest stay serial" each touch many test files (the first rewrites fixtures and assertions, the second is mostly one use … async: line or one # Serial: comment per file).

LiveView modules that stay serial because of global state or the settings row (handover to #540): member_live/vereinfacht_sync_notice_test (mutates OS env), member_live/index_sort_atom_property_test (VM-wide atom count), member_live/index_groups_performance_test (global telemetry handler), member_live/index_filter_input_hardening_logging_test (captures logs), member_live/index_boolean_filter_performance_test (wall-clock budget), member_field_live/index_component_test, member_live/form_error_handling_test and join_request_live/show_test (update the settings row); unchanged: global_settings_live_test, global_settings_live_config_test, join_live_test, join_live_email_failure_test, auth/link_oidc_account_live_logging_test, member_live/index/setup_test.

closes #611 ## Description of the implemented changes The changes were: - [ ] Bugfixing - [ ] New Feature - [ ] Breaking Change - [ ] Refactoring Run the LiveView test files in parallel: most of the remaining serial block of the fast suite were LiveView test modules whose only serial reason was "flip deferred". They are now hardened against other tests' rows and host load and run async; every module that stays serial names its actual reason. ## What has been changed? - State that the async log-capture guard covers the capture_log tag forms - Move the wall-clock and log-capture tests into their own serial modules - Keep LiveView tests independent of other tests' rows and of host load - Run the LiveView test modules async and name why the rest stay serial - Record the suite times after the LiveView flips and the async pitfalls they surfaced ## Definition of Done ### Code Quality - [ ] No new technical depths - [x] Linting passed - [x] Documentation is added were needed ### Accessibility - [ ] New elements are properly defined with html-tags - [ ] Colour contrast follows WCAG criteria - [ ] Aria labels are added when needed - [ ] Everything is accessible by keyboard - [ ] Tab-Order is comprehensible - [ ] All interactive elements have a visible focus ### Testing - [x] Tests for new code are written - [x] All tests pass - [ ] axe-core dev tools show no critical or major issues ## Additional Notes Fast suite (`just test-fast`) on the same 4 vCPU host, seeds 101/202/303, no other test run on the host, medians: | | Before | After | Δ | |---|---|---|---| | Total | 227.4 s | 162.8 s | −64.6 s (−28 %) | | Async | 55.9 s | 74.6 s | +18.7 s | | Sync | 169.6 s | 88.2 s | −81.4 s | 30 test modules moved from serial to async (74.8 s of serial time). Every flipped module passed `--repeat-until-failure 5` on its own and five seeded full-suite runs per batch, with no Postgres deadlock and no lock wait of 1 s or more. Under parallel load from another worktree on the same host the suite takes 180–215 s. Modules that stay serial after the async proof: `index_sort_atom_property_test` (asserts the VM-wide atom count), `index_value_control_test`, `index_bulk_scope_test`, `index_filter_chips_test`, `index_filter_input_hardening_test` (`assert_patch` timed out under heavy host load; the hardening is kept), and `member_field_live/index_component_test` (a lock wait on the settings row; settings isolation is #540's scope). Re-flipping the four `assert_patch` modules with a longer patch timeout is a follow-up. Deliberate deviation: six overview test modules of the same `assert_patch` class stay async with a 1 s patch timeout instead of going back to serial; they never failed themselves and held five full runs under heavy load. No product code changes; the diff is limited to `test/` and `docs/`. Larger commits: "Keep LiveView tests independent of other tests' rows and of host load" and "Run the LiveView test modules async and name why the rest stay serial" each touch many test files (the first rewrites fixtures and assertions, the second is mostly one `use … async:` line or one `# Serial:` comment per file). LiveView modules that stay serial because of global state or the settings row (handover to #540): `member_live/vereinfacht_sync_notice_test` (mutates OS env), `member_live/index_sort_atom_property_test` (VM-wide atom count), `member_live/index_groups_performance_test` (global telemetry handler), `member_live/index_filter_input_hardening_logging_test` (captures logs), `member_live/index_boolean_filter_performance_test` (wall-clock budget), `member_field_live/index_component_test`, `member_live/form_error_handling_test` and `join_request_live/show_test` (update the settings row); unchanged: `global_settings_live_test`, `global_settings_live_config_test`, `join_live_test`, `join_live_email_failure_test`, `auth/link_oidc_account_live_logging_test`, `member_live/index/setup_test`.

simon added 5 commits

2026-10-20 23:23:17 +02:00

docs(test): state that the async log-capture guard covers the capture_log tag forms a405f60f4a

test(member): move the wall-clock and log-capture tests into their own serial modules f2452418fc

test(live): keep LiveView tests independent of other tests' rows and of host load daf75dd21d

test(live): run the LiveView test modules async and name why the rest stay serial 6235b341eb

Modules that failed the async proof stay serial with their reason: the sort atom property asserts the VM-wide atom count, four overview modules hit assert_patch timeouts under heavy host load, and the member field table showed a settings row lock wait.

docs(test): record the suite times after the LiveView flips and the async pitfalls they surfaced

continuous-integration/drone/push Build is passing

Details

continuous-integration/drone/promote/production Build is passing

Details

805fafc163

simon merged commit b407fb1272 into main

2026-10-20 23:37:17 +02:00

simon deleted branch issue/mitgliederverwaltung-611

2026-10-20 23:37:17 +02:00

simon referenced this pull request from a commit

2026-10-20 23:37:17 +02:00